基于贝叶斯分类器的脉象自动识别方法

传统脉诊依靠医生按压腕部挠动脉脉搏进行脉象识别,具有很强的主观性和模糊性,其准确性与可靠性依赖于医生个人的主观感觉与经验积累,缺乏客观和量化的诊断指标.针对脉象信号复杂性以及脉象特征与脉象类别之间非线性等特点,提出一种基于贝叶斯分类器的脉象自动识别方法,并据此建立脉象定量诊断模型.首先,提取脉象信号的特征参数,创建脉象特征参数-脉象类别数据库,采用少数类合成过采样技术SMOTE结合Tomek links的方法,对数据库进行均衡,使不同的脉象类别具有大致相同的样本;然后基于均衡后的数据库学习贝叶斯网络结构,将得到的马尔可夫毯选择为特征集合并作为贝叶斯分类器的输入,创建脉象信号与类别之间的映射关系模型.通过创建的脉象样本数据库和交叉验证方法,对所提出的方法进行验证.结果表明:所提出的方法可有效识别脉象类型,对于脉位、脉率和脉律的预测准确率都超过90%,是一种有效的脉象定量诊断方法.

Automatic Pulse Recognition Method Based on Bayesian Classifier

Pulse diagnosis is one of the most important examinations in traditional Chinese medicine (TCM). Due to the subjectivity and fuzziness of pulse diagnosis in TCM, quantitative methods are demanded. In view of the complexity of pulse signals and the nonlinear relationships between pulse parameters and pulse types, a new pulse signal recognition method was proposed and pulse quantitative diagnostic models were built. First, the characteristic parameters of pulse signals were computed and a database containing characteristic parameters of pulse signals and pulse types was built. Second, the dataset was balanced using synthetic minority over-sampling technique (SMOTE) and Tomek links method, which made different pulse types have approximately equal sample number. Third, a Bayesian network structure was learned from the balanced database and the Markov blanket of pulse type in the structure was selected as feature set and taken as input of a Bayesian classifier. Fourth, the mapping relationships between pulse signals and pulse types were constructed. The proposed methodology was testified by the established pulse sample database and cross validation method. It was showed that most pulse types had high predictive accuracy and the predictive accuracies of depth, frequency and rhythm were higher than 90%. The proposed method was effective and suitable for pulse signal classification.